ばか！バカ！　馬鹿ー月！

I have been curious about how machine translation is effectively done for a while now. I remember a while back, I tried to use a OCR program on light novels only to find it miserably fail at recognizing correctly. What process do people who machine translate typically do?

Setting aside the issue of whether machine translation is ever effective at all (some would argue that it is still too inaccurate)...

Yes, OCR has its limitations when converting scanned pages into actual text.

This is one reason why people who use machine translation tend to stick to web novels, whose text source is readily available, instead of taking on light novels where one would have to fix all the kanji errors in the OCR'd text first.